Re: [PATCH 2/4] arm64: entry: Always inline helpers used by non-instrumentable paths

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 3 potential
issues to consider:
- [Low] arm64: entry: kasan_hw_tags_enabled() not marked __always_inline
  for noinstr
- [Low] arm64: entry: sme_{enter,exit} helpers call non-noinstr
  test_thread_flag()
- [Low] arm64: entry: #else stubs for MTE and SME helpers not marked
  __always_inline
--

commit 36003a6dfa6a3ccb4d813e94b64d8956098d7ce8
Author: Jisheng Zhang <jszhang@kernel.org>

arm64: entry: Always inline helpers used by non-instrumentable paths

This commit changes several MTE, SME, CPU feature, and entry helpers from
plain inline to __always_inline so they are not emitted out-of-line in
.text when called from noinstr entry and exit paths.

> diff --git a/arch/arm64/include/asm/fpsimd.h b/arch/arm64/include/asm/fpsimd.h
> index a67d5774e6728..157d0551f6279 100644
> --- a/arch/arm64/include/asm/fpsimd.h
> +++ b/arch/arm64/include/asm/fpsimd.h
> @@ -749,14 +749,14 @@ void sme_enable_dvmsync(void);
>  void sme_set_active(void);
>  void sme_clear_active(void);
>  
> -static inline void sme_enter_from_user_mode(void)
> +static __always_inline void sme_enter_from_user_mode(void)
>  {
>  	if (alternative_has_cap_unlikely(ARM64_WORKAROUND_4193714) &&
>  	    test_thread_flag(TIF_SME))
>  		sme_clear_active();
>  }
>  
> -static inline void sme_exit_to_user_mode(void)
> +static __always_inline void sme_exit_to_user_mode(void)
>  {
>  	if (alternative_has_cap_unlikely(ARM64_WORKAROUND_4193714) &&
>  	    test_thread_flag(TIF_SME))

[Severity: Low]
Should sme_enter_from_user_mode() and sme_exit_to_user_mode() use
tif_test_bit(TIF_SME) instead of test_thread_flag(TIF_SME)?

test_thread_flag() expands to test_ti_thread_flag(), which is plain static
inline and calls instrumented test_bit():

include/linux/thread_info.h:test_ti_thread_flag() {
	return test_bit(flag, (unsigned long *)&ti->flags);
}

Inlining these SME helpers across the noinstr EL0 entry and exit paths in
arch/arm64/kernel/entry-common.c allows the compiler to outline
test_ti_thread_flag() into .text under CONFIG_CC_OPTIMIZE_FOR_SIZE=y or emit
bitop instrumentation in .noinstr.text, whereas tif_test_bit() is already
marked __always_inline and avoids bitop instrumentation for noinstr callers.

[ ... ]

> diff --git a/arch/arm64/include/asm/mte.h b/arch/arm64/include/asm/mte.h
> index 83f3b05fc7849..b2bf2e70c9aec 100644
> --- a/arch/arm64/include/asm/mte.h
> +++ b/arch/arm64/include/asm/mte.h
> @@ -222,7 +222,7 @@ static inline bool folio_try_hugetlb_mte_tagging(struct folio *folio)
>  }
>  #endif
>  
> -static inline void mte_disable_tco_entry(struct task_struct *task)
> +static __always_inline void mte_disable_tco_entry(struct task_struct *task)
>  {
>  	if (!system_supports_mte())
>  		return;
> @@ -245,7 +245,7 @@ static inline void mte_disable_tco_entry(struct task_struct *task)
>  #ifdef CONFIG_KASAN_HW_TAGS
>  void mte_check_tfsr_el1(void);
>  
> -static inline void mte_check_tfsr_entry(void)
> +static __always_inline void mte_check_tfsr_entry(void)
>  {
>  	if (!kasan_hw_tags_enabled())
>  		return;

[Severity: Low]
Does kasan_hw_tags_enabled() also need to be marked __always_inline?

mte_disable_tco_entry(), mte_check_tfsr_entry(), and mte_check_tfsr_exit()
all call kasan_hw_tags_enabled(), which is still defined as plain static
inline in include/linux/kasan-enabled.h (even though the underlying
kasan_enabled() helper is already __always_inline). Force-inlining the MTE
entry and exit helpers into arch/arm64/kernel/entry-common.c expands
kasan_hw_tags_enabled() across dozens of noinstr call sites, so GCC with
CONFIG_CC_OPTIMIZE_FOR_SIZE=y and CONFIG_KASAN_HW_TAGS=y can still emit an
out-of-line copy of kasan_hw_tags_enabled() in .text.

> @@ -256,7 +256,7 @@ static inline void mte_check_tfsr_entry(void)
>  	mte_check_tfsr_el1();
>  }
>  
> -static inline void mte_check_tfsr_exit(void)
> +static __always_inline void mte_check_tfsr_exit(void)
>  {
>  	if (!kasan_hw_tags_enabled())
>  		return;

[Severity: Low]
Should the #else stubs for mte_check_tfsr_entry() and mte_check_tfsr_exit()
(when !CONFIG_KASAN_HW_TAGS) and for sme_enter_from_user_mode() and
sme_exit_to_user_mode() in arch/arm64/include/asm/fpsimd.h (when
!CONFIG_ARM64_SME) also be marked __always_inline to match
system_uses_mte_async_or_asymm_mode()?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261107071339.17467-1-jszhang@kernel.org?part=2